Semantic Retrieval of Instructional Videos

A content based retrieval model is presented that mea- sures the semantic relevance between a query and instruc- tional videos. In the feedback loop, users provide feedback by means of partial orders of the retrieved items. The partial order feedback enables the users to order the retrieved items but does not enforce a well defined ordering. The system adapts to the user's partial order feedback by optimizing the correlation between the semantic concepts in the lecture videos. The optimization is formulated as a quadratic pro- gramming problem. If an optimal solution does not exist for the optimization problem, the system uses an approximation algorithm that reduces the number of contradicting orders. Experiments with six users demonstrate the effectiveness of the retrieval system, given limited feedback information.